Millisecond Pulsar

What defines a millisecond pulsar

The operational definition is simple: spin period P_spin below about 30 ms — more than ~33 full rotations per second. That cutoff is somewhat conventional but it captures a population that is genuinely distinct from "normal" young pulsars. MSPs are old, have weak magnetic fields, low period derivatives, very long characteristic ages, and overwhelmingly live in (or were once in) binary systems. They cluster in their own corner of the P-Ṗ diagram: short P, tiny Ṗ, far away from the locus of young pulsars born of recent supernovae.

The fastest known MSP — and the fastest known rotating astrophysical body of any kind — is PSR J1748-2446ad, discovered in 2006 inside the globular cluster Terzan 5. It spins at 1.396 ms, or 716 Hz. At that rate, a point on the equator of a 12 km neutron star moves at roughly a quarter of the speed of light. Push much faster and the centrifugal force at the equator approaches gravitational binding; the star would shed mass. So MSPs are not just fast — they are pressed against the upper limit imposed by neutron-star structure.

Recycling: how you spin up a dead pulsar

A young pulsar is born with a strong magnetic field (~10¹² G) and a period of tens of milliseconds to a second. Over ~10⁷ years it radiates rotational kinetic energy through magnetic-dipole braking and slows, eventually crossing the so-called "death line" beyond which the gap voltage above the magnetic poles is too low to ignite pair cascades. The pulsar goes dark.

If the dead pulsar happens to have a stellar companion, however, evolution can resurrect it. As the companion expands off the main sequence and overflows its Roche lobe, gas falls onto the neutron star through an accretion disk. The gas carries specific angular momentum √(GMr), and the inner edge of the disk transfers that onto the star's surface. Steady mass transfer over 10⁸–10⁹ years can torque the spin period down to milliseconds. Meanwhile, the accreted plasma buries or screens the surface magnetic field down to 10⁸–10⁹ G — three to four orders of magnitude weaker than at birth. The weaker field means a much smaller spin-down torque, which is why MSPs have Ṗ ~ 10⁻²⁰ instead of the 10⁻¹⁵ typical of young pulsars and characteristic ages τ = P/(2Ṗ) of order 10⁹ years.

This "recycling" picture, articulated by Alpar, Cheng, Ruderman and Shaham in 1982, predicts that MSPs should overwhelmingly live in binaries — and they do. The "missing link" between low-mass X-ray binaries (the accreting phase) and MSPs (the radio-emitting end state) was confirmed when a handful of transitional MSPs, such as PSR J1023+0038, were observed to switch back and forth between X-ray accretion and radio pulsar states on month-to-year timescales.

The P-Ṗ diagram and the MSP cluster

The pulsar P-Ṗ diagram is to pulsar astronomy what the Hertzsprung-Russell diagram is to stellar astronomy. Each pulsar is plotted at its measured spin period P and period derivative Ṗ. Lines of constant characteristic age (P/2Ṗ), constant inferred surface field (B ∝ √(PṖ)), and the death line all run as diagonals. Young pulsars (~10⁴–10⁶ yr) cluster in the upper-middle: P ~ 0.1–1 s, Ṗ ~ 10⁻¹⁴.

Millisecond pulsars occupy a distinct lower-left clump: P from 1.4 to 30 ms and Ṗ between 10⁻²² and 10⁻¹⁹. Their inferred fields cluster at 10⁸–10⁹ G; their characteristic ages reach 10⁹–10¹⁰ years. The cleanest evidence that MSPs are not simply an extreme tail of young pulsars but a separate population produced by a different evolutionary channel is precisely this gap in the P-Ṗ plane between the two clusters.

Class	P_spin	Ṗ	B (inferred)	Char. age	Typical companion
Young pulsar	0.03 – 1 s	10⁻¹⁵ – 10⁻¹³	10¹¹ – 10¹³ G	10⁴ – 10⁶ yr	Isolated / runaway
Mildly recycled	10 – 100 ms	10⁻¹⁸ – 10⁻¹⁶	10⁹ – 10¹⁰ G	10⁷ – 10⁸ yr	Massive WD / NS
Millisecond pulsar	1.4 – 30 ms	10⁻²² – 10⁻¹⁹	10⁸ – 10⁹ G	10⁹ – 10¹⁰ yr	He WD, CO WD, sub-stellar
Magnetar	2 – 12 s	10⁻¹² – 10⁻¹⁰	10¹⁴ – 10¹⁵ G	10³ – 10⁵ yr	Isolated

Backer 1982 and PSR B1937+21

The first MSP, PSR B1937+21, was discovered in 1982 by Donald C. Backer, Shrinivas R. Kulkarni and collaborators using the 305 m Arecibo telescope. Its period — 1.5578064688197945 ms — was a shock. The fastest pulsar known up to that point spun at 33 ms (the Crab); B1937+21 was 20 times faster. The discovery paper appeared as Backer et al., Nature 300, 615 (1982).

Two surprises came with it. First, B1937+21 was isolated, not in a binary — the companion that recycled it had been ablated away. (Such "spider" systems — black widows with substellar companions, redbacks with main-sequence stars — are now a known sub-class.) Second, the period derivative was tiny: Ṗ ~ 10⁻¹⁹. The combination implied a weak field (~4 × 10⁸ G) and a characteristic age of ~2 × 10⁸ yr, immediately solidifying the recycling picture proposed by Alpar et al. the same year. B1937+21 also turned out to be exceptionally stable, becoming the prototype for using MSPs as precision clocks.

Globular clusters: MSP factories

About a third of all known MSPs live in globular clusters, which is wildly over-represented relative to the cluster contribution to the galaxy's stellar mass (~0.1%). The reason is dynamical: globular cluster cores are stellar-density traffic jams in which exchange interactions efficiently swap neutron stars into close binaries, then those binaries are perturbed by further encounters into configurations that lead to Roche-lobe overflow and recycling.

Terzan 5, a metal-rich cluster ~6.9 kpc away in the bulge, hosts more than 40 known MSPs — the largest single MSP factory we have. It also harbours PSR J1748-2446ad (the 1.40 ms record holder), the most massive MSP known, PSR J1748-2446J at ~2.74 M☉ (though this is contested), and a population of relativistic-binary MSPs ideal for tests of GR. The next-richest cluster, 47 Tucanae, has ~30 MSPs. Most cluster MSPs are isolated or have helium white dwarf companions; the ablating black-widow population is also enriched.

Timing precision: rivalling atomic clocks

The pulse profile of an MSP, averaged over thousands of rotations, is extraordinarily stable in shape. Each observed arrival time at the telescope (a "TOA") can be modelled with a deterministic ephemeris that includes:

• Spin period P and derivative Ṗ (plus higher-order terms if needed)
• Astrometric parameters: position, proper motion, parallax
• Time-variable dispersion measure DM(t) along the line of sight
• For binaries: five Keplerian + several post-Keplerian parameters
• Earth's motion in the solar-system barycentre (using JPL ephemerides)
• Atomic-clock time-transfer corrections (TT(BIPM) → TCB)

What remains after subtracting the best-fit model is the timing residual. For the best MSPs — PSR J1909-3744, PSR J0437-4715, PSR J1713+0747 — rms residuals over a decade of monitoring are at the level of tens of nanoseconds. That precision is comparable to the long-term stability of laboratory caesium fountains, and on certain timescales the pulsars are arguably more stable than terrestrial clocks. The Bureau International des Poids et Mesures has explored using pulsar timescales as a check on atomic time.

Pulsar timing arrays

If MSPs are clocks, an array of MSPs distributed across the sky is a galaxy-scale interferometer. A passing gravitational wave of nanohertz frequency stretches and squeezes the spacetime metric between us and each pulsar. The fractional change in the photon travel time imprints a small additional residual on the TOAs. For a single pulsar, that residual is degenerate with countless other things (clock errors, ephemeris errors, intrinsic spin wandering). The trick is that a real GW background produces a very specific correlation between pulsar pairs — the Hellings-Downs curve, derived by Hellings and Downs in 1983 — that depends only on the angular separation θ between the pulsars:

ζ(θ) = (1/2) − (x/4) + (3/2) x ln x ,    x = (1 − cos θ)/2

This curve crosses through ζ(0) = 0.5 (pulsars in the same direction, fully correlated), drops to a minimum near θ ≈ 80°, and returns toward zero at antipodal angles. The Hellings-Downs signature is the fingerprint that distinguishes a stochastic GW background from monopolar (clock-error) or dipolar (ephemeris) systematics, which produce different angular correlations.

Three regional collaborations have been timing MSPs for two decades to mount this experiment:

Array	Telescope(s)	Pulsars timed	Started
NANOGrav (North America)	Arecibo (until 2020), Green Bank, VLA	~67	2007
EPTA (Europe)	Effelsberg, Westerbork, Lovell, Nançay, SRT	~25	2005
PPTA (Australia)	Parkes 64 m	~25	2003
InPTA (India)	uGMRT	~10	2015
CPTA (China)	FAST 500 m	~50	2019
IPTA (combined)	All of the above	~80	2008

NANOGrav 2023 — Hellings-Downs to 4σ

In June 2023, NANOGrav, EPTA+InPTA, PPTA and CPTA simultaneously released results showing evidence for a Hellings-Downs-correlated stochastic gravitational-wave background at nanohertz frequencies. NANOGrav's headline analysis of 15 years of data on 68 pulsars reported the correlation at the 3.5–4σ level depending on assumptions and statistics. EPTA+InPTA reported ~3σ, PPTA ~2σ, and a coordinated IPTA cross-check is ongoing.

The most natural astrophysical source is the cosmic population of inspiralling supermassive black-hole binaries (SMBHBs), the remnants of galaxy mergers throughout the universe. Their incoherent sum produces a stochastic background with characteristic strain h_c(f) ∝ f⁻²ᐟ³, a prediction first written down by Phinney in 2001. The observed background has a slightly redder tilt than the simplest SMBHB model, leaving room for individual loud sources or for cosmological contributions: first-order phase transitions, cosmic-string networks, primordial GWs from inflation, or a small contribution from a yet-to-be-resolved continuous source.

The detection opens a new band in GW astronomy. LIGO/Virgo observe at 10 Hz – kHz (stellar-mass mergers, milliseconds to seconds). LISA (launching ~2035) will observe at mHz (massive black-hole mergers, hours). PTAs are now opening the nHz band (galactic-mass SMBH inspirals, years to decades). Each band probes a different astrophysical population.

Testing general relativity with binary MSPs

A binary MSP with a compact companion is a precision laboratory for strong-field gravity. The basic idea is to measure as many post-Keplerian (PK) parameters as you can. Each PK parameter is, in GR, a known function of the two component masses (and only the two masses). If you measure three PK parameters, GR predicts the third given the first two — any disagreement is a violation. Measuring more than three over-determines the system and tests GR independently in each direction.

The most useful PK parameters are:

• ω̇ — periastron advance (analogue of Mercury's perihelion precession)
• γ — combined transverse-Doppler + gravitational redshift
• Ṗ_b — orbital period decay from GW emission (Hulse-Taylor 1975 prize-winning effect)
• r, s — Shapiro delay range and shape: the extra arrival-time delay when pulses pass near the companion's mass
• Ω̇_SO — spin-orbit precession of the pulsar spin axis

The double pulsar PSR J0737-3039 A/B, discovered in 2003, is unique: both neutron stars are radio pulsars, and the system is edge-on so Shapiro delay is huge. Two decades of timing have measured seven PK parameters; GR passes every test at the 0.05% level or better. PSR J1614-2230 and PSR J0740+6620 give Shapiro-delay mass measurements (~1.9 M☉ and ~2.08 M☉ respectively) that exclude soft equations of state. PSR J0337+1715 is a triple system — an MSP plus two white dwarfs — that tests the strong equivalence principle to ~2 × 10⁻⁶.

Probing the neutron-star equation of state

Nuclear matter at the densities inside a neutron star — a few times ρ_nuc = 2.7 × 10¹⁴ g/cm³ — is not accessible in any laboratory. The pressure-density relation P(ρ), the "equation of state" (EoS), is uncertain by factors of order unity at the relevant densities and depends on poorly constrained physics: nucleon-nucleon three-body forces, hyperon thresholds, possible quark-matter phases, pion or kaon condensates.

What we can measure is the masses and radii of real neutron stars. The Tolman-Oppenheimer-Volkoff equations turn any candidate EoS into a unique mass-radius curve and a maximum mass M_max. If observations show a neutron star with M > M_cand_max, that EoS is ruled out.

MSP mass measurements have been decisive. PSR J1614-2230 (Demorest et al. 2010, 1.97 M☉) was the first ironclad measurement above 1.9 M☉. PSR J0348+0432 (Antoniadis et al. 2013, 2.01 ± 0.04 M☉) and especially PSR J0740+6620 (Cromartie et al. 2020, 2.08 ± 0.07 M☉) cemented the case. Any EoS that cannot support 2.08 M☉ — and that excludes many models with significant hyperon or strange-quark softening — is dead.

Combine this with radius measurements: NICER (the Neutron-star Interior Composition Explorer on the ISS) has constrained R ≈ 12.4 km for PSR J0740+6620 by modelling the pulsed X-ray hot-spot waveform. Together, the mass-radius constraints are tightening on a roughly intermediate-stiffness EoS, consistent with the GW170817 tidal-deformability constraints from LIGO. The MSPs are still doing most of the work on the high-mass end.

Worked example: spin-up by accretion

How long does it take to spin a neutron star up to milliseconds? The torque per unit mass at the accretion radius is the specific angular momentum √(GM r_in). For accretion onto a neutron star at the magnetospheric radius r_mag (where magnetic pressure balances ram pressure), one has

N = Ṁ √(GM r_mag)   (spin-up torque)

and the equilibrium spin period reached when the magnetosphere co-rotates with the disk is

P_eq ≈ 1.9 ms × (B/10⁹ G)⁶ᐟ⁷ × (Ṁ/Ṁ_Edd)⁻³ᐟ⁷ × (M/1.4 M☉)⁻⁵ᐟ⁷

Plug in a recycled-pulsar field B = 10⁸ G and Ṁ near Eddington: P_eq drops below 1 ms — explaining why MSPs cluster near 1.4–10 ms rather than spreading uniformly. The total angular momentum needed to spin a 1.4 M☉ neutron star (I ≈ 10⁴⁵ g cm²) from rest to 1 ms (Ω = 2π × 10³ rad/s) is

L = I Ω ≈ 6 × 10⁴⁸ g cm²/s

Delivered at the Eddington rate Ṁ_Edd ≈ 1.4 × 10¹⁸ g/s with specific L ~ √(GM × 10⁶ cm) ~ 10¹⁶ cm²/s, the accreted mass needed is ΔM ~ 6 × 10⁴⁸ / 10¹⁶ ≈ 6 × 10³² g ≈ 0.3 M☉. The accretion timescale is therefore

t_recycle ~ ΔM / Ṁ ~ 6 × 10³² / (1.4 × 10¹⁸) ≈ 4 × 10¹⁴ s ≈ 10⁷ yr   (Eddington)

Most real recycling proceeds well below Eddington, stretching this to 10⁸–10⁹ yr, which sits comfortably with the observed binary evolution timescales. The order-of-magnitude conclusion is robust: accreting a few tenths of a solar mass is enough to spin a dead pulsar back to milliseconds.

Sub-classes and oddities

• Black widows. MSPs in tight orbits (P_b < 10 hr) with sub-stellar (~0.02–0.05 M☉) companions that the pulsar wind is actively ablating. PSR B1957+20 is the prototype. Some companions show eclipses of the radio signal.
• Redbacks. Similar but with non-degenerate main-sequence companions (~0.2–0.4 M☉). Show longer-duration eclipses from intra-binary plasma. PSR J1023+0038 transitioned between accretion-powered and rotation-powered states in 2013, confirming the recycling picture in real time.
• Transitional MSPs (tMSPs). Switch back and forth between LMXB accretion and radio pulsar states. PSR J1023+0038, PSR J1227-4853, and XSS J12270-4859 are the three confirmed members.
• Sub-millisecond candidates. Several searches have hunted pulsars below 1 ms; none confirmed yet. The 716 Hz of PSR J1748-2446ad is uncomfortably close to predictions of the centrifugal break-up frequency for stiffer EoS, which is part of why sub-ms candidates would be a strong EoS constraint.
• Isolated MSPs. About 20% of MSPs in the galactic field are isolated, presumed to have ablated or merged with their recycling companion. In globular clusters the fraction is higher, due to dynamical disruption.

Common pitfalls

• Conflating MSPs with all fast pulsars. The Crab pulsar rotates at 33 ms — not an MSP. The 30 ms cutoff is operational, capturing the recycled-binary population; merely being "fast" doesn't make you an MSP.
• Reading too much into a single timing residual. A pulsar shows residuals from many sources: dispersion-measure variations, magnetospheric jitter, scintillation, ephemeris errors. Treating a single anomaly as "evidence for new physics" is a perennial mistake. The whole point of timing arrays is the cross-pulsar correlation pattern.
• Forgetting that Hellings-Downs is the signature. A red common-spectrum process in all pulsars is not, by itself, a GW background — could be unmodelled clock or ephemeris errors. Only the Hellings-Downs angular correlation between pairs is the GW fingerprint. The 2020 NANOGrav 12.5-yr paper saw a common-spectrum process; the 2023 paper added the Hellings-Downs detection.
• Treating the recycling picture as universal. A small fraction of MSPs may have formed through alternative channels — accretion-induced collapse of a massive white dwarf, or unusual electron-capture supernovae — that bypass the LMXB phase. The binary statistics, eccentricity distributions, and Galactic latitude distribution all point overwhelmingly to standard recycling, but exceptions exist.
• Confusing characteristic age with true age. τ = P/(2Ṗ) is exact only if P_0 ≪ P and the spin-down is constant-B dipole. For recycled MSPs born near their current P, τ vastly overestimates the true age. White-dwarf cooling ages of the companion are typically more reliable.